On the HDT with the Tree Representation for Large RDFs on GPU

Abstract

Nowadays, semantic web technology depends on interexchange and integration of RDF data for various aspects of each social communication. The searching for possible answer to the related topics across the open data source obviously becomes a massive task. In this research, we study the RDF query with parallel processing on the GPU. In particular, in this paper, we consider the compact data representation for the RDFs which can enable importing more data to the GPUs memory to enable parallel search in the GPU. The key idea is the use of compressed data type such as HDT before going the search on GPUs. Loading the HDT file to the GPUs straightforwardly and perform searching may not be the good solutions. Thus, this work presents the tree representation from the HDT data which can com-pact the HDT triples, ease the GPU memory transfer, and enable the GPU parallel search. With the HDT representation, the size is reduced from the original RDF about 10%-30%. Together with the tree array representation, we can reduce the redundant terms from in HDT triples by 30%-50% for the test cases.